patchwork-python

Python port of the R patchwork package.

patchwork lets you compose multiple ggplot figures into a single layout using arithmetic-style operators — same idea here, same operators, on top of ggplot2-python:

import patchwork  # installs the | / + - dunders on GGPlot
from ggplot2_py import ggplot, aes, geom_point

p1 = ggplot(df) + geom_point(aes("x", "y"))
p2 = ggplot(df) + geom_point(aes("x", "z"))

p1 | p2          # side-by-side
p1 / p2          # stacked
(p1 | p2) / p3   # nested

Where it fits in the Bio-Babel R-port stack

patchwork-python    <-- this repo: composing ggplots
        |
        +-- ggplot2-python      grammar of graphics
        +-- rgrid-python        viewport / unit / Cairo backend
        +-- gtable-python       layout-aware grob tables
        +-- scales-python       scale transformations
        +-- great_tables        (optional) backend for wrap_table()

Install

pip install patchwork-python
pip install "patchwork-python[tables]"   # adds great_tables for wrap_table()

Figure size

Jupyter's _repr_png_ has no current graphics device, so render size lives on the plot object. Defaults are 7.0 × 5.0 in @ 150 dpi (patchwork/_display.py).

pw = p1 | p2
pw.fig_width = 12
pw.fig_height = 8
pw.fig_dpi = 200
pw   # renders at 12×8 @ 200 dpi

Patchwork._repr_png_ resolves hints in this order: self → self.plot → defaults. So p.fig_width = 12; p | q propagates without re-setting on the wrapper. A bare Patch reads from its inner plot only (Patch.__slots__ blocks per-instance attrs).

R contrast: none. R routes size through the active graphics device (png(), knitr chunk options, RStudio plot pane). The fig_* protocol is a Python-only carry-over from ggplot2_py.GGPlot.

Documentation

Tutorials are mirrored to docs/tutorials/ and rendered via mkdocs:

pip install -e ".[docs]"
mkdocs serve